Berikut adalah README.txt yang menarik dan informatif untuk model yang telah Anda unggah ke Kaggle Model Hub:
Fine-tuned DistilBERT Model for Indonesian Text Classification
Overview
This repository contains a fine-tuned version of the DistilBERT model (based on cahya/distilbert-base-indonesian) for Indonesian text classification. The model is trained to classify text into eight distinct categories, including politics, socio-cultural, defense and security, ideology, economy, natural resources, demography, and geography.
Dataset
The dataset used for training the model underwent significant augmentation and balancing to address class imbalance issues. Below are the details of the dataset before and after augmentation:
Before Augmentation
Category | Count |
---|---|
Politik | 2972 |
Sosial Budaya | 587 |
Pertahanan dan Keamanan | 400 |
Ideologi | 400 |
Ekonomi | 367 |
Sumber Daya Alam | 192 |
Demografi | 62 |
Geografi | 20 |
After Augmentation
Category | Count |
---|---|
Politik | 2969 |
Demografi | 427 |
Sosial Budaya | 422 |
Ideologi | 343 |
Pertahanan dan Keamanan | 331 |
Ekonomi | 309 |
Sumber Daya Alam | 156 |
Geografi | 133 |
Label Encoding
Encoded | Label |
---|---|
0 | Demografi |
1 | Ekonomi |
2 | Geografi |
3 | Ideologi |
4 | Pertahanan dan Keamanan |
5 | Politik |
6 | Sosial Budaya |
7 | Sumber Daya Alam |
Data Split
The dataset was split into training and testing sets with an 85:15 ratio.
- Train Size: 4326 samples
- Test Size: 764 samples
Model Training
The model was trained for 4 epochs, achieving the following results:
Epoch | Train Loss | Train Accuracy |
---|---|---|
1 | 1.0240 | 0.6766 |
2 | 0.5615 | 0.8220 |
3 | 0.3270 | 0.9014 |
4 | 0.1759 | 0.9481 |
Training Completion
- Test Loss: 0.7948
- Test Accuracy: 0.7687
- Test Balanced Accuracy: 0.7001
Model Evaluation
The model was evaluated using precision, recall, and F1 scores, with the following results:
- Precision Score: 0.7714
- Recall Score: 0.7696
- F1 Score: 0.7697
Classification Report
Category | Precision | Recall | F1-Score | Support |
---|---|---|---|---|
Demografi | 0.94 | 0.91 | 0.92 | 64 |
Ekonomi | 0.67 | 0.72 | 0.69 | 46 |
Geografi | 0.95 | 0.95 | 0.95 | 20 |
Ideologi | 0.71 | 0.56 | 0.62 | 52 |
Pertahanan dan Keamanan | 0.69 | 0.66 | 0.67 | 50 |
Politik | 0.84 | 0.85 | 0.84 | 446 |
Sosial Budaya | 0.38 | 0.40 | 0.39 | 63 |
Sumber Daya Alam | 0.50 | 0.57 | 0.53 | 23 |
- Accuracy: 0.7696
- Balanced Accuracy: 0.7001
- Macro Avg Precision: 0.71
- Macro Avg Recall: 0.70
- Macro Avg F1-Score: 0.70
- Weighted Avg Precision: 0.77
- Weighted Avg Recall: 0.77
- Weighted Avg F1-Score: 0.77
Usage
To use this model, you can load it using the Hugging Face Transformers library:
# Use a pipeline as a high-level helper
from transformers import pipeline
pipe = pipeline("text-classification", model="Rendika/Trained-DistilBERT-Indonesia-Presidential-Election-Balanced-Dataset")
# Load model directly
from transformers import AutoTokenizer, AutoModelForSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("Rendika/Trained-DistilBERT-Indonesia-Presidential-Election-Balanced-Dataset")
model = AutoModelForSequenceClassification.from_pretrained("Rendika/Trained-DistilBERT-Indonesia-Presidential-Election-Balanced-Dataset")
Conclusion
This fine-tuned DistilBERT model for Indonesian text classification demonstrates robust performance across various categories. The augmentation and balancing of the dataset have contributed significantly to the model's ability to generalize well on the test set.
Feel free to use this model for your Indonesian text classification tasks, and don't hesitate to reach out if you have any questions or feedback.
- Downloads last month
- 15